Disease Classification and Biomarker Discovery Using ECG Data.

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Cardiac disease classification and biomarker discovery are crucial in medical research.
• Electrocardiograms (ECGs) offer a low-cost, noninvasive method for heart disease screening.
• The proliferation of personal ECG monitors necessitates efficient data analysis algorithms.

Purpose of the Study:

• To develop an efficient and interpretable cardiac disease classification procedure.
• To explore novel feature extraction methods for ECG data.
• To compare the performance of different classification algorithms.

Main Methods:

• Characterized ECG data distributions using sample quantiles, addressing skewed, heavy-tailed, or multimodal patterns.
• Employed Principal Component Analysis (PCA) for dimension reduction.
• Compared three classifiers: Stepwise Discriminant Analysis (SDA), Support Vector Machine (SVM), and LASSO logistic regression.

Main Results:

• Stepwise Discriminant Analysis (SDA) applied to PCA-dimension-reduced features demonstrated superior stability and effectiveness.
• Achieved high diagnostic performance metrics: 89.68% sensitivity, 84.62% specificity, and 88.52% accuracy.
• Sample quantiles proved more effective than sample means for characterizing ECG data distributions.

Conclusions:

• The developed SDA procedure using PCA-reduced features offers a robust and accurate method for cardiac disease classification.
• This approach provides an efficient and interpretable tool for analyzing large ECG datasets.
• The findings highlight the utility of quantile-based feature extraction in ECG analysis.